What changes when you stop grading the destination and start considering the journey?
A few years ago at Holy Innocents’ Episcopal School, we ran what I thought was a genuinely good AI lesson. Students were engaged, the room had energy, and the activity did exactly what we designed it to do. Afterward, I met with the teacher and heard the sentence that’s since been played back in some version a dozen times:
“The activity was cool, and the kids seemed to enjoy it, but I don’t know how to grade it, and I’m honestly not sure they learned anything.”
That sentence redirected how we approach working with AI at HIES. We had been doing what most schools were doing: writing guidelines for student and teacher use and searching for tools worth putting in front of kids. What that teacher named was a problem no set of guidelines or policy statement was going to solve.
New Rules for Driving With AI
Two things convinced us that a policy could not be the center of the work. The first was simple: rules were expiring faster than we could write them. Every few weeks, an AI tool gains a new capability or gets embedded into a platform we already license, and the policy we finalized last month no longer describes the situation our students are actually facing. Or someone brings us a use case the policy never accounted for, with a lot of gray area. We were trying to write a policy for a program that had already changed.
The second came from some of our own Challenge Success survey data. The results of the survey pointed to a gap between achievement and engagement. Students were earning strong grades, but many reported feeling less connected to the material than the grades would suggest. In conversations with teachers, too many students had a hard time explaining the “why” behind what they were learning. The good grades-thin engagement dynamic had everything to do with why AI was about to be such a problem for us.
Students who are already “doing” school rather than learning are the ones most likely to hand the work to a tool. If students do not understand or value the purpose of the work, it becomes easier to hand an assignment off to a tool that can provide the answer in a matter of seconds.
Once you combine the speed of AI change with student perception, the question stops being about what should be the rules for AI and instead becomes something deeper: What does learning look like now, and how would we actually know it happened?
Teacher Observations: Mind the Gap
Our first instinct was to build a wall: lock down devices; move everything into the classroom; keep AI out of student work while encouraging teachers to learn how to use it for planning and differentiation. But, with this approach, teachers noticed a gap between the work and the student. Assignments completed outside of class were turned in but did not sound like the student who wrote it – not all the time and not every student, but increasingly.
When teachers questioned students about their work, some couldn’t talk about their own paper. Even though sometimes they had written it, students allowed AI to revise their own voice out of it. These assignments failed to show teachers the students’ current level of understanding to then be able to provide meaningful feedback for improvement. This is much more of a missed learning opportunity than an integrity problem.
Plenty of teachers respond to AI by going analog, with notebooks, handwritten assignments, and requiring completion of projects in the classroom. I understand the appeal, and it does solve part of the problem. Yet, this leads to class time spent on work we used to assign outside of class, meaning time to dig deeper is gone. Going analog does not move us anywhere new.
What learning are we sacrificing when everything has to happen in the room? Plus, this approach does not address the moment a student walks off our campus, with AI in their pocket and inside every platform they already use. We cannot control access, but we can design learning that holds up.
Turning On the “GPS”
What we built at HIES, and are now running with an opt-in cohort of teachers, is a four-phase learning cycle grounded in how the brain actually learns. We run this cycle for each learning objective rather than once per unit, beginning with the student’s own thinking.
Before instruction, students retrieve what they already know and commit to an initial position, which becomes their baseline. Then comes the learning phase, which looks like traditional teaching: direct instruction, notes, labs, practice problems, discussion, video. Students receive formative feedback that guides them toward accuracy, but they are not yet being graded for accuracy.
There are checkpoints where students return to their baseline and show how it has changed. At the end of the learning phase, students restate their current understanding in light of everything learned together. The last phase is transfer: the same learning objective, new context, different variables, performed on their own.
Professor Carol Ann Tomlinson taught us that students take different paths to the same destination (differentiation). I like to use the metaphor of a road trip. Not everyone starts in the same place. Before you can give directions, you have to turn on the GPS, which I consider the baseline. For the checkpoints, we ask about progress made, roadblocks faced, and detours taken. For the transfer task, we say, “You have arrived; now tell me about the trip. Then drive it again on your own with the GPS off.”
This structure is where AI is allowed to live; the cycle is about learning rather than the artifact. Students still make things. The question is not how polished the product is but how did project completion help with understanding. If a student uses AI to help with the writing or to get the assignment done, we are not looking at what they submitted but what was learned.
Shifting Gears: Moving Where the Grade Lives
The biggest mindset shift for teachers with this four-phase learning model is how to grade and where we “put the points.” In cohort classrooms using this framework, approximately three-fourths of a student’s grade on a single learning objective comes from engagement with the learning process. The rest is derived from the transfer task.
Engagement does not simply mean compliance or participation. Students have to show genuine effort and authentic thinking against a set of criteria, which they either meet or do not meet yet. If they do not meet the criteria, students revise and resubmit, right up until the transfer task, at which point we ask them to drive on their own.
While we care about process, too often educators put heavier weight on the final product. Those products were intended to give teachers an idea of what a student knows, but now the product can be manufactured. Once AI can generate a polished output in seconds, that inconsistency stops being philosophical and becomes a practical problem.
Yet, this is not really a new problem, which is worth reminding faculty who feel ambushed. When I taught math, I had students turn in flawless homework, yet it was a tutor at home who walked them through every problem and told them how to write it down. The gradebook looked terrific until test day. AI is the same problem at scale and at speed, now reaching every student, not just those whose families could afford tutors. There is also no embarrassment in using AI: asking another person to do your work costs you something, but asking a chatbot costs you nothing.
Our college counseling office at HIES noticed one unexpected benefit of shifting grading. Their early read is that grading is easier to explain to universities because, within this framework, an A, B, or C each mean something specific. A traditional C could reflect a student who did every assignment and failed every test, but it could also represent a student who did no assignments but aced the assessments. Within this revised framework, a C tells you a student has been engaged in class but cannot yet transfer the learning on their own. An A or B in those same courses tells you they can make the transfer – that students have been thinking deeply about the content and reflecting on their own learning along the way.
How to Travel: Taking the Stairs or the Escalator?
Picture a steep hill with a set of stairs on one side and an escalator running next to it. Which one do you choose? We’re wired to take the easier path, and there’s no shame in that. Now picture a student at 11:30 PM, home from a game or performance, with an assignment due in the morning. The escalator is tempting.
What I hope faculty consider is that we already know what good teaching looks like. We know how students’ brains learn. None of that has changed, but now there’s a tool that can take a student deeper and further than they could have gone alone. That same tool can also bypass the learning entirely. Our job is not to tear out the escalator; it’s to get students walking up it. They still build muscle, moving faster than they could on the stairs.
This imagery has done much to align our faculty. Underlying is the principle that there’s a benefit and a loss in every use of AI. I encourage schools to name both. A history teacher and a math teacher can disagree completely about whether a chatbot belongs in a particular assignment and still agree that AI can provide both a benefit and a cost if used. Everyone can also agree that the first thinking should always be the student’s, which is likened to them taking the first steps to get moving.
Another powerful reframing is the teacher’s role. AI is not replacing teachers, but it does provide the chance to “duplicate” them. With anywhere from 15 to 30 students in a room, providing feedback is different from one-on-one. But a custom GPT or other tool loaded with a teacher’s rubric and course material can give a student something useful while they wait their turn. Teachers still have the conversation, but students are not stuck while waiting; they keep walking up the escalator.
Once teachers shift that mindset, students start to see things differently too. The question is not whether they got there, because that can be faked, but how they got there, and, to know this, we have to change what we look for. Every math teacher already understands this: students must show their work to solve math problems because, if the work is correct, the answer takes care of itself. If all I see is an answer, I have no way of knowing how students reached the answer or if they copied a neighbor’s work. Looking at the process can be applied to every school subject.
Throughout this framework, students are building real AI fluency. They are learning what the tool is good for, what it costs them, and when to keep their own hands on the wheel. But that only works if there is friction in the use. If a student can hand the whole thing over with a finished answer, they have bypassed the learning entirely. AI is the most convincing illusion of understanding a student has ever had access to. It will hand them a page that looks exactly like “knowing.” Our job as educators is to keep asking for the part that cannot be handed over.
Getting Everyone on the Bus: What the Data Show
Our strongest evidence to date is qualitative, stemming from cohort teachers who have taught the same course for years and can tell you precisely how this group differs from previous groups. Faculty report more students speaking up, asking questions, able to explain “why” rather than just “what.”
Two things stand out: first, in cohort classes, our second semester seniors, historically the most “checked out” humans on any campus, stayed active through the end of the year. Second, our middle schoolers in the cohorted classes contribute to discussions, offer their own ideas, and argue more because class stopped being about being right and became about sharing what you think and watching it change. When I surveyed students from the pilot group, they described class as more thinking and understanding rather than superficial learning to take a quiz. When students reach for that vocabulary on their own, something has moved in the right direction.
It takes time to get there, though. Many of our students have gotten very good at “doing” school. They can memorize what a teacher says and reproduce it the way the teacher wants it presented. They are far less comfortable thinking on their own and look for immediate validation instead of trusting themselves. Trusting their own thinking first may be the most valuable thing we can give students as they grow up in an AI-enabled world.
We are still building the quantitative picture. Our plan this year is to look at how cohort students’ engagement responses in surveys compare with those of students in traditional sections, and we are also watching final grades across cohort and non-cohort courses.
What we are building at Holy Innocents’ is a solid attempt at a real problem. It is a way, but it is not the only way. Any school that starts asking these questions honestly is going to build something shaped by its own faculty and its own students, as it should. The framework is not the point; knowing how students got there is.
Teacher Paul Matthews uses the analogy “ChatGPT tennis.” The teacher uses AI to write the assessment. The student uses AI to complete it. The teacher uses AI to grade it. We have just removed the two people who were supposed to be at the center. The answer is not to put AI into everything we do. Learning is transformation, not production. Our job is to make that transformation visible and to show how a student’s thinking actually moved down the road.